Different Aggergates For Different Hierarchies?
May 27, 2008
Hi,
Nwebie question:
is it possible to have a fact that will behave diffeerntly as the user
is viewing it from different hierarchies? say i got creatin measures
of a product inventory. i got a location hierarchy, as as user is
'moving up' in it (say,up by city, region, state, country...)i want to
do a sum. but i also got time hierarchy, and as user begins to group
by weeks, months, years... a sum makes no sense..i want LastNonEmpty
or something like that. Is it possible?
thanks!
View 9 Replies
Feb 6, 2005
As promised, question #2... ;)
As I mentioned in my previous post, my "real" case is a cube with 79 dimensions, most of which virtual, have been added for convenience.
Think for instance about a time dimension... Wouldn't it be nice to get a matrix with years horizontally, months vertically and displaying say the number of order you had for each cell in the resulting grid.
Ok, maybe you can do this with MDX but not in Excel, unless you create virtual dimensions for the Year and Month levels.
That's all good, but as it is, in my real case, I end up with four date dimensions for which I have to provide:
YQMD (Year, Quarter, Month & Day) hirarchized dimension
YWD (Year, Week, Day)
YQMD (Fiscal calendar)
Year
Quarter
Month
Week
Day (as 1, .. 31)
Week Day (for periodicity analysis over a week's time)
Date (as individual day - this is the backbone for the virtual dimensions)
It turns out this makes 40 dimensions by itself. For the sake of it, I grouped them by 4 hierarchies, although I've seen no specific functionality off of this in the data browser or Excel, so it really seems to be only for "show".
Now in my previous post I explained how I "spread" my session count to calculate a conversion rate. Given the number of dimensions I have (very high segmentation at the order level, very limited segmentation at the session/visit level), this means my calculated cell formula looks like this (hold your breath, it's ugly):
SubCube: {[Measures].[Session Count]}, DESCENDANTS([Business Unit].[All]), DESCENDANTS([Channel Source].[All]), DESCENDANTS([Currency].[All]), DESCENDANTS([FO Product Segment].[All]), DESCENDANTS([FO Quarter].[All]), DESCENDANTS([FO Value Segment].[All]), DESCENDANTS([Is Cancelled].[All]), DESCENDANTS([Is Customer Care].[All]), DESCENDANTS([Is Design Services].[All]), DESCENDANTS([Is Direct].[All]), DESCENDANTS([Is Estimated].[All]), DESCENDANTS([Is Holiday Order].[All]), DESCENDANTS([Is New].[All]), DESCENDANTS([Is Sales Services].[All]), DESCENDANTS([Language].[All]), DESCENDANTS([Payment].[All]), DESCENDANTS([Product Segment].[All]), DESCENDANTS([Value Segment].[All]), DESCENDANTS([FO Channel].[All]), DESCENDANTS([Reg Channel].[All]), DESCENDANTS([FCD].[Date].[All]), DESCENDANTS([FCD].[Day].[All]), DESCENDANTS([FCD].[Mth].[All]), DESCENDANTS([FCD].[Qtr].[All]), DESCENDANTS([FCD].[Wk].[All]), DESCENDANTS([FCD].[WkDay].[All]), DESCENDANTS([FCD].[Year].[All]), DESCENDANTS([FCD].[Fiscal].[All]), DESCENDANTS([FCD].[YMQD].[All]), DESCENDANTS([FCD].[YWD].[All]), DESCENDANTS([FRD].[Date].[All]), DESCENDANTS([FRD].[Day].[All]), DESCENDANTS([FRD].[Mth].[All]), DESCENDANTS([FRD].[Qtr].[All]), DESCENDANTS([FRD].[Wk].[All]), DESCENDANTS([FRD].[WkDay].[All]), DESCENDANTS([FRD].[Year].[All]), DESCENDANTS([FRD].[Fiscal].[All]), DESCENDANTS([FRD].[YQMD].[All]), DESCENDANTS([FRD].[YWD].[All]), DESCENDANTS([FSD].[Date].[All]), DESCENDANTS([FSD].[Day].[All]), DESCENDANTS([FSD].[Mth].[All]), DESCENDANTS([FSD].[Qtr].[All]), DESCENDANTS([FSD].[Wk].[All]), DESCENDANTS([FSD].[WkDay].[All]), DESCENDANTS([FSD].[Year].[All]), DESCENDANTS([FSD].[Fiscal].[All]), DESCENDANTS([FSD].[YQMD].[All]), DESCENDANTS([FSD].[YWD].[All])
Caculation Value: CalculationPassValue(([Business Unit].[All],[Channel Source].[All],[Currency].[All],[FO Product Segment].[All],[FO Quarter].[All],[FO Value Segment].[All],[Is Cancelled].[All],[Is Customer Care].[All],[Is Design Services].[All],[Is Direct].[All],[Is Estimated].[All],[Is Holiday Order].[All],[Is New].[All],[Is Sales Services].[All],[Language].[All],[Payment].[All],[Product Segment].[All],[Value Segment].[All],[FO Channel].[All],[Reg Channel].[All],[FCD].[Date].[All],[FCD].[Day].[All],[FCD].[Mth].[All],[FCD].[Qtr].[All],[FCD].[Wk].[All],[FCD].[WkDay].[All],[FCD].[Year].[All],[FCD].[Fiscal].[All],[FCD].[YMQD].[All],[FCD].[YWD].[All],[FRD].[Date].[All],[FRD].[Day].[All],[FRD].[Mth].[All],[FRD].[Qtr].[All],[FRD].[Wk].[All],[FRD].[WkDay].[All],[FRD].[Year].[All],[FRD].[Fiscal].[All],[FRD].[YQMD].[All],[FRD].[YWD].[All],[FSD].[Date].[All],[FSD].[Day].[All],[FSD].[Mth].[All],[FSD].[Qtr].[All],[FSD].[Wk].[All],[FSD].[WkDay].[All],[FSD].[Year].[All],[FSD].[Fiscal].[All],[FSD].[YQMD].[All],[FSD].[YWD].[All],[Measures].[Session Count]), 0)
If you read all this, you can see already the cryptic dimension names like "FRD", "FSD" and so forth... that's because with the real names ("First Refund Date", "First Ship Date", the query processor errored out... visibly there is a limit in the size of the formulas you can post!
Is there no other way to achieve this result? Basically I mean to say: if the session count is not defined at your level along this dimension, go to the root of the dimension to get the value there, this along a slew of dimensions, many of which are inherently dependent because of the usage of virtual dimensions (therefore if I wish to go to the root of my "First Refund Date" for instance, I wish to do so along all sub-dimensions... Heck, as far as I'm concerned this is conceptually only ONE dimension, just with various views upon it...
Using hierachies I was sort of hoping for the ability to have something like:
[FRD].[All Hierachies].[All]
Am I just asking for too much or do I just not know (quite probable) the magic keyword that can do this?
Where this is becoming quite critical is that I actually have a calculated cell that goes as follows (abridged):
SubCube: similar as before
Calculation: [Measures].[Order Count]*CalculationPassValue((...set of all un-tied dimension roots..., [Measures].[Distributed Marketing Cost])/CalculationPassValue((...set of all un-tied dimension roots..., [Measures].[Order Count])
Now the purpose of this is to distribute external costs at the order level. In short, say that you know you spent $10,000 globally promoting a specific group of websites in commission money (you pay $1 for each order). This formula allows me to get that a specific website, with 20 orders incurred an additional marketing cost of $20. That's actually the object of my next thread's question (spreading a multiplication through the aggregations)
For the purpose of this thread I am just concerned about the size of my formulas. Renaming the dimensions seemed to "buy" me some margin and I was actually quite surprised to find that the formula still fitted in, and works, but it is only a matter of time until I have to add more dimensions and the whole thing blows up in my face. Adittionally this is obviously not pleasant to look at and maintain.
Any better solution?
Thanks,
Seb
View 1 Replies
View Related
Jan 24, 2004
Hey all. I have a query where I am basically querying an organizational chart. The table storing this information is basically a two column table of parent/child pairs. So you might have:
parent | child
1 | 2
2 | 3
2 | 4
3 | 6
3 | 7
4 | 8
5 | 9
8 | 10
8 | 11
Querying this table should return all child columns that flow up to the top, so querying 1 would return 1, 2, 3, 4, 6, 7, 8, 10, 11.
I used a sample from MSDN (http://www.sqljunkies.com/Article/D7CAED46-CCAC-4FF7-B528-B2E9A274B71C.scuk) that does just this, except that when querying a value that returns more than 1000 records - I am experiencing way too long response times. The MSDN sample uses temp tables and inserts values into a temp table as it moves through the records and finds a match.
Anyone have an ideas on another way to accomplish the same thing? This is an important part of my security model as users that login should only have access to data that falls within their parent/child heirarchy.
Thanks.
Jon
View 20 Replies
View Related
Oct 31, 2007
i guess really i'm looking for opinions and / or experiences of modelling organisational hierarchies.... anyone?
i'm in the process of creating a logical model (very early stages) of a new datawarehouse. The current OLTP schema stores it as a self referencing key (i.e. parent_group_id). The performance problems involved in aggregating to different levels of the hierarchy causes no end of complaints from customers, as i'm sure you can imagine. So, now we are modelling a datawarehouse for their reporting requirements, i obviously want it to be better
so far i've considered...
1) keeping it as it is
2) allocating a 'node' id and storing the left / right nodes in the hierarchy tree. we've used before when reporting over the OLTP system with report developers building it 'on the fly' as part of their report.
(better explanation than mine... http://www.dbpd.com/vault/9811/kamfn.shtml )
3) an intermediate table storing its 'position' in the hierarchy? not really played with this yet but mr kimball seems to like it
see... http://www.dbmsmag.com/9809d05.html
4) denormalising it completely. i.e. level_1, level_2, level_3, Level4.... for every fact? the advantage of this is that it makes aggregation at any level really easy. the problem is that each of our customers has a different organisation hierarchy, with a different number of levels, and given the OLTP schema there is no limit to the number of levels in a hierarchy
so... thoughts / experiences anyone?
Em
View 2 Replies
View Related
Jul 20, 2005
Is there a good approach to modelling many heterogeneous entity typeswith that have some attributes in common?Say I have entities "employees" which share some attibutes (e.g.firstname, lastname, dateofbirth) but some subsets of employees (e.g.physicians, janitors, nurses, ambulance drivers) may have additionalattributes that do not apply to all employees. Physicians may haveattributes specialty and date of board certification, ambulancedrivers may have a drivers license id, janitors may havepreferredbroomtype and so on.There are many employee subtypes and more can be dynamically addedafter the application is deployed so it's obviously no good to keepadding attributes to the employees table because most attributes willbe NULL (since janitors are never doctors at the same time).The only solution I found for this is a generalization hiearchy whereyou have the employee table with all generic attributes and then youadd tables for each new employee subtype as necessary. The subtypetables share the primary key of the employee table. The employee tablehas a "discriminator" field that allows you to figure out whichsubtype table to load for a particular entity.This solution does not seem to scale since for each value of"discriminator" I need to perform a join with a different table. Whatif I need to retrieve 1,000 employees at once?Is that possible to obtain a single ResultSet with one SQL statementSQL?Or do you I need to iterate look at the discriminator and thenperform the appropriate join? If this kind of iteration is necessarythen obviously this generalization hierarchy approach does not work inpracticesince it would be painfully slow.Is there a better approach to modelling these kind of heterogeneousentities with shared attributes that does not involve creating a tablefor each new employee type or having sparce tables (mostly filled withNULLS)I guess another approach would be to use name/value pairs but thatwould make reporting really ugly.Seems like a very common problem. Any ideas? Is this a fundamentallimitation of SQL?Thanks!- robert
View 13 Replies
View Related
Dec 17, 2006
Hi,
I am currently trying to develop an application which would help in retrieving data from cubes (Microsoft Analysis Services)! The user would not be accessing the Business Intelligence Studio, etc. but would be viewing the data from a custom made application developed in VB.Net2005.
While implementing this, I want to populate the drop-down-list in the VB(.Net) Form, by retrieving the various hierarchies in the dimensions of the cubes (along with dimensions if possible). This should be done dynamically and in real-time!
Can you please help in implementing this? Any code/method, etc would be highly appreciated!
Thanks in advance.
Best wishes!
(Software : SQL Server 2005 Enterprise Edition (with Analysis Services and BI Studio), VS.Net 2005 Enterprise Edition, ADOMD.Net)
View 1 Replies
View Related
